xref: /linux/tools/testing/selftests/net/gro.c (revision 621cde16e49b3ecf7d59a8106a20aaebfb4a59a9)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * This testsuite provides conformance testing for GRO coalescing.
4  *
5  * Test cases:
6  * 1.data
7  *  Data packets of the same size and same header setup with correct
8  *  sequence numbers coalesce. The one exception being the last data
9  *  packet coalesced: it can be smaller than the rest and coalesced
10  *  as long as it is in the same flow.
11  * 2.ack
12  *  Pure ACK does not coalesce.
13  * 3.flags
14  *  Specific test cases: no packets with PSH, SYN, URG, RST set will
15  *  be coalesced.
16  * 4.tcp
17  *  Packets with incorrect checksum, non-consecutive seqno and
18  *  different TCP header options shouldn't coalesce. Nit: given that
19  *  some extension headers have paddings, such as timestamp, headers
20  *  that are padding differently would not be coalesced.
21  * 5.ip:
22  *  Packets with different (ECN, TTL, TOS) header, ip options or
23  *  ip fragments (ipv6) shouldn't coalesce.
24  * 6.large:
25  *  Packets larger than GRO_MAX_SIZE packets shouldn't coalesce.
26  *
27  * MSS is defined as 4096 - header because if it is too small
28  * (i.e. 1500 MTU - header), it will result in many packets,
29  * increasing the "large" test case's flakiness. This is because
30  * due to time sensitivity in the coalescing window, the receiver
31  * may not coalesce all of the packets.
32  *
33  * Note the timing issue applies to all of the test cases, so some
34  * flakiness is to be expected.
35  *
36  */
37 
38 #define _GNU_SOURCE
39 
40 #include <arpa/inet.h>
41 #include <errno.h>
42 #include <error.h>
43 #include <getopt.h>
44 #include <linux/filter.h>
45 #include <linux/if_packet.h>
46 #include <linux/ipv6.h>
47 #include <net/ethernet.h>
48 #include <net/if.h>
49 #include <netinet/in.h>
50 #include <netinet/ip.h>
51 #include <netinet/ip6.h>
52 #include <netinet/tcp.h>
53 #include <stdbool.h>
54 #include <stddef.h>
55 #include <stdio.h>
56 #include <stdarg.h>
57 #include <string.h>
58 #include <unistd.h>
59 
60 #include "../kselftest.h"
61 
62 #define DPORT 8000
63 #define SPORT 1500
64 #define PAYLOAD_LEN 100
65 #define NUM_PACKETS 4
66 #define START_SEQ 100
67 #define START_ACK 100
68 #define ETH_P_NONE 0
69 #define TOTAL_HDR_LEN (ETH_HLEN + sizeof(struct ipv6hdr) + sizeof(struct tcphdr))
70 #define MSS (4096 - sizeof(struct tcphdr) - sizeof(struct ipv6hdr))
71 #define MAX_PAYLOAD (IP_MAXPACKET - sizeof(struct tcphdr) - sizeof(struct ipv6hdr))
72 #define NUM_LARGE_PKT (MAX_PAYLOAD / MSS)
73 #define MAX_HDR_LEN (ETH_HLEN + sizeof(struct ipv6hdr) + sizeof(struct tcphdr))
74 #define MIN_EXTHDR_SIZE 8
75 #define EXT_PAYLOAD_1 "\x00\x00\x00\x00\x00\x00"
76 #define EXT_PAYLOAD_2 "\x11\x11\x11\x11\x11\x11"
77 
78 #define ipv6_optlen(p)  (((p)->hdrlen+1) << 3) /* calculate IPv6 extension header len */
79 #define BUILD_BUG_ON(condition) ((void)sizeof(char[1 - 2*!!(condition)]))
80 
81 static const char *addr6_src = "fdaa::2";
82 static const char *addr6_dst = "fdaa::1";
83 static const char *addr4_src = "192.168.1.200";
84 static const char *addr4_dst = "192.168.1.100";
85 static int proto = -1;
86 static uint8_t src_mac[ETH_ALEN], dst_mac[ETH_ALEN];
87 static char *testname = "data";
88 static char *ifname = "eth0";
89 static char *smac = "aa:00:00:00:00:02";
90 static char *dmac = "aa:00:00:00:00:01";
91 static bool verbose;
92 static bool tx_socket = true;
93 static int tcp_offset = -1;
94 static int total_hdr_len = -1;
95 static int ethhdr_proto = -1;
96 static const int num_flush_id_cases = 6;
97 
vlog(const char * fmt,...)98 static void vlog(const char *fmt, ...)
99 {
100 	va_list args;
101 
102 	if (verbose) {
103 		va_start(args, fmt);
104 		vfprintf(stderr, fmt, args);
105 		va_end(args);
106 	}
107 }
108 
setup_sock_filter(int fd)109 static void setup_sock_filter(int fd)
110 {
111 	const int dport_off = tcp_offset + offsetof(struct tcphdr, dest);
112 	const int ethproto_off = offsetof(struct ethhdr, h_proto);
113 	int optlen = 0;
114 	int ipproto_off, opt_ipproto_off;
115 	int next_off;
116 
117 	if (proto == PF_INET)
118 		next_off = offsetof(struct iphdr, protocol);
119 	else
120 		next_off = offsetof(struct ipv6hdr, nexthdr);
121 	ipproto_off = ETH_HLEN + next_off;
122 
123 	/* Overridden later if exthdrs are used: */
124 	opt_ipproto_off = ipproto_off;
125 
126 	if (strcmp(testname, "ip") == 0) {
127 		if (proto == PF_INET)
128 			optlen = sizeof(struct ip_timestamp);
129 		else {
130 			BUILD_BUG_ON(sizeof(struct ip6_hbh) > MIN_EXTHDR_SIZE);
131 			BUILD_BUG_ON(sizeof(struct ip6_dest) > MIN_EXTHDR_SIZE);
132 			BUILD_BUG_ON(sizeof(struct ip6_frag) > MIN_EXTHDR_SIZE);
133 
134 			/* same size for HBH and Fragment extension header types */
135 			optlen = MIN_EXTHDR_SIZE;
136 			opt_ipproto_off = ETH_HLEN + sizeof(struct ipv6hdr)
137 				+ offsetof(struct ip6_ext, ip6e_nxt);
138 		}
139 	}
140 
141 	/* this filter validates the following:
142 	 *	- packet is IPv4/IPv6 according to the running test.
143 	 *	- packet is TCP. Also handles the case of one extension header and then TCP.
144 	 *	- checks the packet tcp dport equals to DPORT. Also handles the case of one
145 	 *	  extension header and then TCP.
146 	 */
147 	struct sock_filter filter[] = {
148 			BPF_STMT(BPF_LD  + BPF_H   + BPF_ABS, ethproto_off),
149 			BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, ntohs(ethhdr_proto), 0, 9),
150 			BPF_STMT(BPF_LD  + BPF_B   + BPF_ABS, ipproto_off),
151 			BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, IPPROTO_TCP, 2, 0),
152 			BPF_STMT(BPF_LD  + BPF_B   + BPF_ABS, opt_ipproto_off),
153 			BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, IPPROTO_TCP, 0, 5),
154 			BPF_STMT(BPF_LD  + BPF_H   + BPF_ABS, dport_off),
155 			BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, DPORT, 2, 0),
156 			BPF_STMT(BPF_LD  + BPF_H   + BPF_ABS, dport_off + optlen),
157 			BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, DPORT, 0, 1),
158 			BPF_STMT(BPF_RET + BPF_K, 0xFFFFFFFF),
159 			BPF_STMT(BPF_RET + BPF_K, 0),
160 	};
161 
162 	struct sock_fprog bpf = {
163 		.len = ARRAY_SIZE(filter),
164 		.filter = filter,
165 	};
166 
167 	if (setsockopt(fd, SOL_SOCKET, SO_ATTACH_FILTER, &bpf, sizeof(bpf)) < 0)
168 		error(1, errno, "error setting filter");
169 }
170 
checksum_nofold(void * data,size_t len,uint32_t sum)171 static uint32_t checksum_nofold(void *data, size_t len, uint32_t sum)
172 {
173 	uint16_t *words = data;
174 	int i;
175 
176 	for (i = 0; i < len / 2; i++)
177 		sum += words[i];
178 	if (len & 1)
179 		sum += ((char *)data)[len - 1];
180 	return sum;
181 }
182 
checksum_fold(void * data,size_t len,uint32_t sum)183 static uint16_t checksum_fold(void *data, size_t len, uint32_t sum)
184 {
185 	sum = checksum_nofold(data, len, sum);
186 	while (sum > 0xFFFF)
187 		sum = (sum & 0xFFFF) + (sum >> 16);
188 	return ~sum;
189 }
190 
tcp_checksum(void * buf,int payload_len)191 static uint16_t tcp_checksum(void *buf, int payload_len)
192 {
193 	struct pseudo_header6 {
194 		struct in6_addr saddr;
195 		struct in6_addr daddr;
196 		uint16_t protocol;
197 		uint16_t payload_len;
198 	} ph6;
199 	struct pseudo_header4 {
200 		struct in_addr saddr;
201 		struct in_addr daddr;
202 		uint16_t protocol;
203 		uint16_t payload_len;
204 	} ph4;
205 	uint32_t sum = 0;
206 
207 	if (proto == PF_INET6) {
208 		if (inet_pton(AF_INET6, addr6_src, &ph6.saddr) != 1)
209 			error(1, errno, "inet_pton6 source ip pseudo");
210 		if (inet_pton(AF_INET6, addr6_dst, &ph6.daddr) != 1)
211 			error(1, errno, "inet_pton6 dest ip pseudo");
212 		ph6.protocol = htons(IPPROTO_TCP);
213 		ph6.payload_len = htons(sizeof(struct tcphdr) + payload_len);
214 
215 		sum = checksum_nofold(&ph6, sizeof(ph6), 0);
216 	} else if (proto == PF_INET) {
217 		if (inet_pton(AF_INET, addr4_src, &ph4.saddr) != 1)
218 			error(1, errno, "inet_pton source ip pseudo");
219 		if (inet_pton(AF_INET, addr4_dst, &ph4.daddr) != 1)
220 			error(1, errno, "inet_pton dest ip pseudo");
221 		ph4.protocol = htons(IPPROTO_TCP);
222 		ph4.payload_len = htons(sizeof(struct tcphdr) + payload_len);
223 
224 		sum = checksum_nofold(&ph4, sizeof(ph4), 0);
225 	}
226 
227 	return checksum_fold(buf, sizeof(struct tcphdr) + payload_len, sum);
228 }
229 
read_MAC(uint8_t * mac_addr,char * mac)230 static void read_MAC(uint8_t *mac_addr, char *mac)
231 {
232 	if (sscanf(mac, "%hhx:%hhx:%hhx:%hhx:%hhx:%hhx",
233 		   &mac_addr[0], &mac_addr[1], &mac_addr[2],
234 		   &mac_addr[3], &mac_addr[4], &mac_addr[5]) != 6)
235 		error(1, 0, "sscanf");
236 }
237 
fill_datalinklayer(void * buf)238 static void fill_datalinklayer(void *buf)
239 {
240 	struct ethhdr *eth = buf;
241 
242 	memcpy(eth->h_dest, dst_mac, ETH_ALEN);
243 	memcpy(eth->h_source, src_mac, ETH_ALEN);
244 	eth->h_proto = ethhdr_proto;
245 }
246 
fill_networklayer(void * buf,int payload_len)247 static void fill_networklayer(void *buf, int payload_len)
248 {
249 	struct ipv6hdr *ip6h = buf;
250 	struct iphdr *iph = buf;
251 
252 	if (proto == PF_INET6) {
253 		memset(ip6h, 0, sizeof(*ip6h));
254 
255 		ip6h->version = 6;
256 		ip6h->payload_len = htons(sizeof(struct tcphdr) + payload_len);
257 		ip6h->nexthdr = IPPROTO_TCP;
258 		ip6h->hop_limit = 8;
259 		if (inet_pton(AF_INET6, addr6_src, &ip6h->saddr) != 1)
260 			error(1, errno, "inet_pton source ip6");
261 		if (inet_pton(AF_INET6, addr6_dst, &ip6h->daddr) != 1)
262 			error(1, errno, "inet_pton dest ip6");
263 	} else if (proto == PF_INET) {
264 		memset(iph, 0, sizeof(*iph));
265 
266 		iph->version = 4;
267 		iph->ihl = 5;
268 		iph->ttl = 8;
269 		iph->protocol	= IPPROTO_TCP;
270 		iph->tot_len = htons(sizeof(struct tcphdr) +
271 				payload_len + sizeof(struct iphdr));
272 		iph->frag_off = htons(0x4000); /* DF = 1, MF = 0 */
273 		if (inet_pton(AF_INET, addr4_src, &iph->saddr) != 1)
274 			error(1, errno, "inet_pton source ip");
275 		if (inet_pton(AF_INET, addr4_dst, &iph->daddr) != 1)
276 			error(1, errno, "inet_pton dest ip");
277 		iph->check = checksum_fold(buf, sizeof(struct iphdr), 0);
278 	}
279 }
280 
fill_transportlayer(void * buf,int seq_offset,int ack_offset,int payload_len,int fin)281 static void fill_transportlayer(void *buf, int seq_offset, int ack_offset,
282 				int payload_len, int fin)
283 {
284 	struct tcphdr *tcph = buf;
285 
286 	memset(tcph, 0, sizeof(*tcph));
287 
288 	tcph->source = htons(SPORT);
289 	tcph->dest = htons(DPORT);
290 	tcph->seq = ntohl(START_SEQ + seq_offset);
291 	tcph->ack_seq = ntohl(START_ACK + ack_offset);
292 	tcph->ack = 1;
293 	tcph->fin = fin;
294 	tcph->doff = 5;
295 	tcph->window = htons(TCP_MAXWIN);
296 	tcph->urg_ptr = 0;
297 	tcph->check = tcp_checksum(tcph, payload_len);
298 }
299 
write_packet(int fd,char * buf,int len,struct sockaddr_ll * daddr)300 static void write_packet(int fd, char *buf, int len, struct sockaddr_ll *daddr)
301 {
302 	int ret = -1;
303 
304 	ret = sendto(fd, buf, len, 0, (struct sockaddr *)daddr, sizeof(*daddr));
305 	if (ret == -1)
306 		error(1, errno, "sendto failure");
307 	if (ret != len)
308 		error(1, errno, "sendto wrong length");
309 }
310 
create_packet(void * buf,int seq_offset,int ack_offset,int payload_len,int fin)311 static void create_packet(void *buf, int seq_offset, int ack_offset,
312 			  int payload_len, int fin)
313 {
314 	memset(buf, 0, total_hdr_len);
315 	memset(buf + total_hdr_len, 'a', payload_len);
316 	fill_transportlayer(buf + tcp_offset, seq_offset, ack_offset,
317 			    payload_len, fin);
318 	fill_networklayer(buf + ETH_HLEN, payload_len);
319 	fill_datalinklayer(buf);
320 }
321 
322 /* send one extra flag, not first and not last pkt */
send_flags(int fd,struct sockaddr_ll * daddr,int psh,int syn,int rst,int urg)323 static void send_flags(int fd, struct sockaddr_ll *daddr, int psh, int syn,
324 		       int rst, int urg)
325 {
326 	static char flag_buf[MAX_HDR_LEN + PAYLOAD_LEN];
327 	static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
328 	int payload_len, pkt_size, flag, i;
329 	struct tcphdr *tcph;
330 
331 	payload_len = PAYLOAD_LEN * psh;
332 	pkt_size = total_hdr_len + payload_len;
333 	flag = NUM_PACKETS / 2;
334 
335 	create_packet(flag_buf, flag * payload_len, 0, payload_len, 0);
336 
337 	tcph = (struct tcphdr *)(flag_buf + tcp_offset);
338 	tcph->psh = psh;
339 	tcph->syn = syn;
340 	tcph->rst = rst;
341 	tcph->urg = urg;
342 	tcph->check = 0;
343 	tcph->check = tcp_checksum(tcph, payload_len);
344 
345 	for (i = 0; i < NUM_PACKETS + 1; i++) {
346 		if (i == flag) {
347 			write_packet(fd, flag_buf, pkt_size, daddr);
348 			continue;
349 		}
350 		create_packet(buf, i * PAYLOAD_LEN, 0, PAYLOAD_LEN, 0);
351 		write_packet(fd, buf, total_hdr_len + PAYLOAD_LEN, daddr);
352 	}
353 }
354 
355 /* Test for data of same length, smaller than previous
356  * and of different lengths
357  */
send_data_pkts(int fd,struct sockaddr_ll * daddr,int payload_len1,int payload_len2)358 static void send_data_pkts(int fd, struct sockaddr_ll *daddr,
359 			   int payload_len1, int payload_len2)
360 {
361 	static char buf[ETH_HLEN + IP_MAXPACKET];
362 
363 	create_packet(buf, 0, 0, payload_len1, 0);
364 	write_packet(fd, buf, total_hdr_len + payload_len1, daddr);
365 	create_packet(buf, payload_len1, 0, payload_len2, 0);
366 	write_packet(fd, buf, total_hdr_len + payload_len2, daddr);
367 }
368 
369 /* If incoming segments make tracked segment length exceed
370  * legal IP datagram length, do not coalesce
371  */
send_large(int fd,struct sockaddr_ll * daddr,int remainder)372 static void send_large(int fd, struct sockaddr_ll *daddr, int remainder)
373 {
374 	static char pkts[NUM_LARGE_PKT][TOTAL_HDR_LEN + MSS];
375 	static char last[TOTAL_HDR_LEN + MSS];
376 	static char new_seg[TOTAL_HDR_LEN + MSS];
377 	int i;
378 
379 	for (i = 0; i < NUM_LARGE_PKT; i++)
380 		create_packet(pkts[i], i * MSS, 0, MSS, 0);
381 	create_packet(last, NUM_LARGE_PKT * MSS, 0, remainder, 0);
382 	create_packet(new_seg, (NUM_LARGE_PKT + 1) * MSS, 0, remainder, 0);
383 
384 	for (i = 0; i < NUM_LARGE_PKT; i++)
385 		write_packet(fd, pkts[i], total_hdr_len + MSS, daddr);
386 	write_packet(fd, last, total_hdr_len + remainder, daddr);
387 	write_packet(fd, new_seg, total_hdr_len + remainder, daddr);
388 }
389 
390 /* Pure acks and dup acks don't coalesce */
send_ack(int fd,struct sockaddr_ll * daddr)391 static void send_ack(int fd, struct sockaddr_ll *daddr)
392 {
393 	static char buf[MAX_HDR_LEN];
394 
395 	create_packet(buf, 0, 0, 0, 0);
396 	write_packet(fd, buf, total_hdr_len, daddr);
397 	write_packet(fd, buf, total_hdr_len, daddr);
398 	create_packet(buf, 0, 1, 0, 0);
399 	write_packet(fd, buf, total_hdr_len, daddr);
400 }
401 
recompute_packet(char * buf,char * no_ext,int extlen)402 static void recompute_packet(char *buf, char *no_ext, int extlen)
403 {
404 	struct tcphdr *tcphdr = (struct tcphdr *)(buf + tcp_offset);
405 	struct ipv6hdr *ip6h = (struct ipv6hdr *)(buf + ETH_HLEN);
406 	struct iphdr *iph = (struct iphdr *)(buf + ETH_HLEN);
407 
408 	memmove(buf, no_ext, total_hdr_len);
409 	memmove(buf + total_hdr_len + extlen,
410 		no_ext + total_hdr_len, PAYLOAD_LEN);
411 
412 	tcphdr->doff = tcphdr->doff + (extlen / 4);
413 	tcphdr->check = 0;
414 	tcphdr->check = tcp_checksum(tcphdr, PAYLOAD_LEN + extlen);
415 	if (proto == PF_INET) {
416 		iph->tot_len = htons(ntohs(iph->tot_len) + extlen);
417 		iph->check = 0;
418 		iph->check = checksum_fold(iph, sizeof(struct iphdr), 0);
419 	} else {
420 		ip6h->payload_len = htons(ntohs(ip6h->payload_len) + extlen);
421 	}
422 }
423 
tcp_write_options(char * buf,int kind,int ts)424 static void tcp_write_options(char *buf, int kind, int ts)
425 {
426 	struct tcp_option_ts {
427 		uint8_t kind;
428 		uint8_t len;
429 		uint32_t tsval;
430 		uint32_t tsecr;
431 	} *opt_ts = (void *)buf;
432 	struct tcp_option_window {
433 		uint8_t kind;
434 		uint8_t len;
435 		uint8_t shift;
436 	} *opt_window = (void *)buf;
437 
438 	switch (kind) {
439 	case TCPOPT_NOP:
440 		buf[0] = TCPOPT_NOP;
441 		break;
442 	case TCPOPT_WINDOW:
443 		memset(opt_window, 0, sizeof(struct tcp_option_window));
444 		opt_window->kind = TCPOPT_WINDOW;
445 		opt_window->len = TCPOLEN_WINDOW;
446 		opt_window->shift = 0;
447 		break;
448 	case TCPOPT_TIMESTAMP:
449 		memset(opt_ts, 0, sizeof(struct tcp_option_ts));
450 		opt_ts->kind = TCPOPT_TIMESTAMP;
451 		opt_ts->len = TCPOLEN_TIMESTAMP;
452 		opt_ts->tsval = ts;
453 		opt_ts->tsecr = 0;
454 		break;
455 	default:
456 		error(1, 0, "unimplemented TCP option");
457 		break;
458 	}
459 }
460 
461 /* TCP with options is always a permutation of {TS, NOP, NOP}.
462  * Implement different orders to verify coalescing stops.
463  */
add_standard_tcp_options(char * buf,char * no_ext,int ts,int order)464 static void add_standard_tcp_options(char *buf, char *no_ext, int ts, int order)
465 {
466 	switch (order) {
467 	case 0:
468 		tcp_write_options(buf + total_hdr_len, TCPOPT_NOP, 0);
469 		tcp_write_options(buf + total_hdr_len + 1, TCPOPT_NOP, 0);
470 		tcp_write_options(buf + total_hdr_len + 2 /* two NOP opts */,
471 				  TCPOPT_TIMESTAMP, ts);
472 		break;
473 	case 1:
474 		tcp_write_options(buf + total_hdr_len, TCPOPT_NOP, 0);
475 		tcp_write_options(buf + total_hdr_len + 1,
476 				  TCPOPT_TIMESTAMP, ts);
477 		tcp_write_options(buf + total_hdr_len + 1 + TCPOLEN_TIMESTAMP,
478 				  TCPOPT_NOP, 0);
479 		break;
480 	case 2:
481 		tcp_write_options(buf + total_hdr_len, TCPOPT_TIMESTAMP, ts);
482 		tcp_write_options(buf + total_hdr_len + TCPOLEN_TIMESTAMP + 1,
483 				  TCPOPT_NOP, 0);
484 		tcp_write_options(buf + total_hdr_len + TCPOLEN_TIMESTAMP + 2,
485 				  TCPOPT_NOP, 0);
486 		break;
487 	default:
488 		error(1, 0, "unknown order");
489 		break;
490 	}
491 	recompute_packet(buf, no_ext, TCPOLEN_TSTAMP_APPA);
492 }
493 
494 /* Packets with invalid checksum don't coalesce. */
send_changed_checksum(int fd,struct sockaddr_ll * daddr)495 static void send_changed_checksum(int fd, struct sockaddr_ll *daddr)
496 {
497 	static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
498 	struct tcphdr *tcph = (struct tcphdr *)(buf + tcp_offset);
499 	int pkt_size = total_hdr_len + PAYLOAD_LEN;
500 
501 	create_packet(buf, 0, 0, PAYLOAD_LEN, 0);
502 	write_packet(fd, buf, pkt_size, daddr);
503 
504 	create_packet(buf, PAYLOAD_LEN, 0, PAYLOAD_LEN, 0);
505 	tcph->check = tcph->check - 1;
506 	write_packet(fd, buf, pkt_size, daddr);
507 }
508 
509  /* Packets with non-consecutive sequence number don't coalesce.*/
send_changed_seq(int fd,struct sockaddr_ll * daddr)510 static void send_changed_seq(int fd, struct sockaddr_ll *daddr)
511 {
512 	static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
513 	struct tcphdr *tcph = (struct tcphdr *)(buf + tcp_offset);
514 	int pkt_size = total_hdr_len + PAYLOAD_LEN;
515 
516 	create_packet(buf, 0, 0, PAYLOAD_LEN, 0);
517 	write_packet(fd, buf, pkt_size, daddr);
518 
519 	create_packet(buf, PAYLOAD_LEN, 0, PAYLOAD_LEN, 0);
520 	tcph->seq = ntohl(htonl(tcph->seq) + 1);
521 	tcph->check = 0;
522 	tcph->check = tcp_checksum(tcph, PAYLOAD_LEN);
523 	write_packet(fd, buf, pkt_size, daddr);
524 }
525 
526  /* Packet with different timestamp option or different timestamps
527   * don't coalesce.
528   */
send_changed_ts(int fd,struct sockaddr_ll * daddr)529 static void send_changed_ts(int fd, struct sockaddr_ll *daddr)
530 {
531 	static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
532 	static char extpkt[sizeof(buf) + TCPOLEN_TSTAMP_APPA];
533 	int pkt_size = total_hdr_len + PAYLOAD_LEN + TCPOLEN_TSTAMP_APPA;
534 
535 	create_packet(buf, 0, 0, PAYLOAD_LEN, 0);
536 	add_standard_tcp_options(extpkt, buf, 0, 0);
537 	write_packet(fd, extpkt, pkt_size, daddr);
538 
539 	create_packet(buf, PAYLOAD_LEN, 0, PAYLOAD_LEN, 0);
540 	add_standard_tcp_options(extpkt, buf, 0, 0);
541 	write_packet(fd, extpkt, pkt_size, daddr);
542 
543 	create_packet(buf, PAYLOAD_LEN * 2, 0, PAYLOAD_LEN, 0);
544 	add_standard_tcp_options(extpkt, buf, 100, 0);
545 	write_packet(fd, extpkt, pkt_size, daddr);
546 
547 	create_packet(buf, PAYLOAD_LEN * 3, 0, PAYLOAD_LEN, 0);
548 	add_standard_tcp_options(extpkt, buf, 100, 1);
549 	write_packet(fd, extpkt, pkt_size, daddr);
550 
551 	create_packet(buf, PAYLOAD_LEN * 4, 0, PAYLOAD_LEN, 0);
552 	add_standard_tcp_options(extpkt, buf, 100, 2);
553 	write_packet(fd, extpkt, pkt_size, daddr);
554 }
555 
556 /* Packet with different tcp options don't coalesce. */
send_diff_opt(int fd,struct sockaddr_ll * daddr)557 static void send_diff_opt(int fd, struct sockaddr_ll *daddr)
558 {
559 	static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
560 	static char extpkt1[sizeof(buf) + TCPOLEN_TSTAMP_APPA];
561 	static char extpkt2[sizeof(buf) + TCPOLEN_MAXSEG];
562 	int extpkt1_size = total_hdr_len + PAYLOAD_LEN + TCPOLEN_TSTAMP_APPA;
563 	int extpkt2_size = total_hdr_len + PAYLOAD_LEN + TCPOLEN_MAXSEG;
564 
565 	create_packet(buf, 0, 0, PAYLOAD_LEN, 0);
566 	add_standard_tcp_options(extpkt1, buf, 0, 0);
567 	write_packet(fd, extpkt1, extpkt1_size, daddr);
568 
569 	create_packet(buf, PAYLOAD_LEN, 0, PAYLOAD_LEN, 0);
570 	add_standard_tcp_options(extpkt1, buf, 0, 0);
571 	write_packet(fd, extpkt1, extpkt1_size, daddr);
572 
573 	create_packet(buf, PAYLOAD_LEN * 2, 0, PAYLOAD_LEN, 0);
574 	tcp_write_options(extpkt2 + MAX_HDR_LEN, TCPOPT_NOP, 0);
575 	tcp_write_options(extpkt2 + MAX_HDR_LEN + 1, TCPOPT_WINDOW, 0);
576 	recompute_packet(extpkt2, buf, TCPOLEN_WINDOW + 1);
577 	write_packet(fd, extpkt2, extpkt2_size, daddr);
578 }
579 
add_ipv4_ts_option(void * buf,void * optpkt)580 static void add_ipv4_ts_option(void *buf, void *optpkt)
581 {
582 	struct ip_timestamp *ts = (struct ip_timestamp *)(optpkt + tcp_offset);
583 	int optlen = sizeof(struct ip_timestamp);
584 	struct iphdr *iph;
585 
586 	if (optlen % 4)
587 		error(1, 0, "ipv4 timestamp length is not a multiple of 4B");
588 
589 	ts->ipt_code = IPOPT_TS;
590 	ts->ipt_len = optlen;
591 	ts->ipt_ptr = 5;
592 	ts->ipt_flg = IPOPT_TS_TSONLY;
593 
594 	memcpy(optpkt, buf, tcp_offset);
595 	memcpy(optpkt + tcp_offset + optlen, buf + tcp_offset,
596 	       sizeof(struct tcphdr) + PAYLOAD_LEN);
597 
598 	iph = (struct iphdr *)(optpkt + ETH_HLEN);
599 	iph->ihl = 5 + (optlen / 4);
600 	iph->tot_len = htons(ntohs(iph->tot_len) + optlen);
601 	iph->check = 0;
602 	iph->check = checksum_fold(iph, sizeof(struct iphdr) + optlen, 0);
603 }
604 
add_ipv6_exthdr(void * buf,void * optpkt,__u8 exthdr_type,char * ext_payload)605 static void add_ipv6_exthdr(void *buf, void *optpkt, __u8 exthdr_type, char *ext_payload)
606 {
607 	struct ipv6_opt_hdr *exthdr = (struct ipv6_opt_hdr *)(optpkt + tcp_offset);
608 	struct ipv6hdr *iph = (struct ipv6hdr *)(optpkt + ETH_HLEN);
609 	char *exthdr_payload_start = (char *)(exthdr + 1);
610 
611 	exthdr->hdrlen = 0;
612 	exthdr->nexthdr = IPPROTO_TCP;
613 
614 	memcpy(exthdr_payload_start, ext_payload, MIN_EXTHDR_SIZE - sizeof(*exthdr));
615 
616 	memcpy(optpkt, buf, tcp_offset);
617 	memcpy(optpkt + tcp_offset + MIN_EXTHDR_SIZE, buf + tcp_offset,
618 		sizeof(struct tcphdr) + PAYLOAD_LEN);
619 
620 	iph->nexthdr = exthdr_type;
621 	iph->payload_len = htons(ntohs(iph->payload_len) + MIN_EXTHDR_SIZE);
622 }
623 
fix_ip4_checksum(struct iphdr * iph)624 static void fix_ip4_checksum(struct iphdr *iph)
625 {
626 	iph->check = 0;
627 	iph->check = checksum_fold(iph, sizeof(struct iphdr), 0);
628 }
629 
send_flush_id_case(int fd,struct sockaddr_ll * daddr,int tcase)630 static void send_flush_id_case(int fd, struct sockaddr_ll *daddr, int tcase)
631 {
632 	static char buf1[MAX_HDR_LEN + PAYLOAD_LEN];
633 	static char buf2[MAX_HDR_LEN + PAYLOAD_LEN];
634 	static char buf3[MAX_HDR_LEN + PAYLOAD_LEN];
635 	bool send_three = false;
636 	struct iphdr *iph1;
637 	struct iphdr *iph2;
638 	struct iphdr *iph3;
639 
640 	iph1 = (struct iphdr *)(buf1 + ETH_HLEN);
641 	iph2 = (struct iphdr *)(buf2 + ETH_HLEN);
642 	iph3 = (struct iphdr *)(buf3 + ETH_HLEN);
643 
644 	create_packet(buf1, 0, 0, PAYLOAD_LEN, 0);
645 	create_packet(buf2, PAYLOAD_LEN, 0, PAYLOAD_LEN, 0);
646 	create_packet(buf3, PAYLOAD_LEN * 2, 0, PAYLOAD_LEN, 0);
647 
648 	switch (tcase) {
649 	case 0: /* DF=1, Incrementing - should coalesce */
650 		iph1->frag_off |= htons(IP_DF);
651 		iph1->id = htons(8);
652 
653 		iph2->frag_off |= htons(IP_DF);
654 		iph2->id = htons(9);
655 		break;
656 
657 	case 1: /* DF=1, Fixed - should coalesce */
658 		iph1->frag_off |= htons(IP_DF);
659 		iph1->id = htons(8);
660 
661 		iph2->frag_off |= htons(IP_DF);
662 		iph2->id = htons(8);
663 		break;
664 
665 	case 2: /* DF=0, Incrementing - should coalesce */
666 		iph1->frag_off &= ~htons(IP_DF);
667 		iph1->id = htons(8);
668 
669 		iph2->frag_off &= ~htons(IP_DF);
670 		iph2->id = htons(9);
671 		break;
672 
673 	case 3: /* DF=0, Fixed - should not coalesce */
674 		iph1->frag_off &= ~htons(IP_DF);
675 		iph1->id = htons(8);
676 
677 		iph2->frag_off &= ~htons(IP_DF);
678 		iph2->id = htons(8);
679 		break;
680 
681 	case 4: /* DF=1, two packets incrementing, and one fixed - should
682 		 * coalesce only the first two packets
683 		 */
684 		iph1->frag_off |= htons(IP_DF);
685 		iph1->id = htons(8);
686 
687 		iph2->frag_off |= htons(IP_DF);
688 		iph2->id = htons(9);
689 
690 		iph3->frag_off |= htons(IP_DF);
691 		iph3->id = htons(9);
692 		send_three = true;
693 		break;
694 
695 	case 5: /* DF=1, two packets fixed, and one incrementing - should
696 		 * coalesce only the first two packets
697 		 */
698 		iph1->frag_off |= htons(IP_DF);
699 		iph1->id = htons(8);
700 
701 		iph2->frag_off |= htons(IP_DF);
702 		iph2->id = htons(8);
703 
704 		iph3->frag_off |= htons(IP_DF);
705 		iph3->id = htons(9);
706 		send_three = true;
707 		break;
708 	}
709 
710 	fix_ip4_checksum(iph1);
711 	fix_ip4_checksum(iph2);
712 	write_packet(fd, buf1, total_hdr_len + PAYLOAD_LEN, daddr);
713 	write_packet(fd, buf2, total_hdr_len + PAYLOAD_LEN, daddr);
714 
715 	if (send_three) {
716 		fix_ip4_checksum(iph3);
717 		write_packet(fd, buf3, total_hdr_len + PAYLOAD_LEN, daddr);
718 	}
719 }
720 
test_flush_id(int fd,struct sockaddr_ll * daddr,char * fin_pkt)721 static void test_flush_id(int fd, struct sockaddr_ll *daddr, char *fin_pkt)
722 {
723 	for (int i = 0; i < num_flush_id_cases; i++) {
724 		sleep(1);
725 		send_flush_id_case(fd, daddr, i);
726 		sleep(1);
727 		write_packet(fd, fin_pkt, total_hdr_len, daddr);
728 	}
729 }
730 
send_ipv6_exthdr(int fd,struct sockaddr_ll * daddr,char * ext_data1,char * ext_data2)731 static void send_ipv6_exthdr(int fd, struct sockaddr_ll *daddr, char *ext_data1, char *ext_data2)
732 {
733 	static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
734 	static char exthdr_pck[sizeof(buf) + MIN_EXTHDR_SIZE];
735 
736 	create_packet(buf, 0, 0, PAYLOAD_LEN, 0);
737 	add_ipv6_exthdr(buf, exthdr_pck, IPPROTO_HOPOPTS, ext_data1);
738 	write_packet(fd, exthdr_pck, total_hdr_len + PAYLOAD_LEN + MIN_EXTHDR_SIZE, daddr);
739 
740 	create_packet(buf, PAYLOAD_LEN * 1, 0, PAYLOAD_LEN, 0);
741 	add_ipv6_exthdr(buf, exthdr_pck, IPPROTO_HOPOPTS, ext_data2);
742 	write_packet(fd, exthdr_pck, total_hdr_len + PAYLOAD_LEN + MIN_EXTHDR_SIZE, daddr);
743 }
744 
745 /* IPv4 options shouldn't coalesce */
send_ip_options(int fd,struct sockaddr_ll * daddr)746 static void send_ip_options(int fd, struct sockaddr_ll *daddr)
747 {
748 	static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
749 	static char optpkt[sizeof(buf) + sizeof(struct ip_timestamp)];
750 	int optlen = sizeof(struct ip_timestamp);
751 	int pkt_size = total_hdr_len + PAYLOAD_LEN + optlen;
752 
753 	create_packet(buf, 0, 0, PAYLOAD_LEN, 0);
754 	write_packet(fd, buf, total_hdr_len + PAYLOAD_LEN, daddr);
755 
756 	create_packet(buf, PAYLOAD_LEN * 1, 0, PAYLOAD_LEN, 0);
757 	add_ipv4_ts_option(buf, optpkt);
758 	write_packet(fd, optpkt, pkt_size, daddr);
759 
760 	create_packet(buf, PAYLOAD_LEN * 2, 0, PAYLOAD_LEN, 0);
761 	write_packet(fd, buf, total_hdr_len + PAYLOAD_LEN, daddr);
762 }
763 
764 /*  IPv4 fragments shouldn't coalesce */
send_fragment4(int fd,struct sockaddr_ll * daddr)765 static void send_fragment4(int fd, struct sockaddr_ll *daddr)
766 {
767 	static char buf[IP_MAXPACKET];
768 	struct iphdr *iph = (struct iphdr *)(buf + ETH_HLEN);
769 	int pkt_size = total_hdr_len + PAYLOAD_LEN;
770 
771 	create_packet(buf, 0, 0, PAYLOAD_LEN, 0);
772 	write_packet(fd, buf, pkt_size, daddr);
773 
774 	/* Once fragmented, packet would retain the total_len.
775 	 * Tcp header is prepared as if rest of data is in follow-up frags,
776 	 * but follow up frags aren't actually sent.
777 	 */
778 	memset(buf + total_hdr_len, 'a', PAYLOAD_LEN * 2);
779 	fill_transportlayer(buf + tcp_offset, PAYLOAD_LEN, 0, PAYLOAD_LEN * 2, 0);
780 	fill_networklayer(buf + ETH_HLEN, PAYLOAD_LEN);
781 	fill_datalinklayer(buf);
782 
783 	iph->frag_off = htons(0x6000); // DF = 1, MF = 1
784 	iph->check = 0;
785 	iph->check = checksum_fold(iph, sizeof(struct iphdr), 0);
786 	write_packet(fd, buf, pkt_size, daddr);
787 }
788 
789 /* IPv4 packets with different ttl don't coalesce.*/
send_changed_ttl(int fd,struct sockaddr_ll * daddr)790 static void send_changed_ttl(int fd, struct sockaddr_ll *daddr)
791 {
792 	int pkt_size = total_hdr_len + PAYLOAD_LEN;
793 	static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
794 	struct iphdr *iph = (struct iphdr *)(buf + ETH_HLEN);
795 
796 	create_packet(buf, 0, 0, PAYLOAD_LEN, 0);
797 	write_packet(fd, buf, pkt_size, daddr);
798 
799 	create_packet(buf, PAYLOAD_LEN, 0, PAYLOAD_LEN, 0);
800 	iph->ttl = 7;
801 	iph->check = 0;
802 	iph->check = checksum_fold(iph, sizeof(struct iphdr), 0);
803 	write_packet(fd, buf, pkt_size, daddr);
804 }
805 
806 /* Packets with different tos don't coalesce.*/
send_changed_tos(int fd,struct sockaddr_ll * daddr)807 static void send_changed_tos(int fd, struct sockaddr_ll *daddr)
808 {
809 	int pkt_size = total_hdr_len + PAYLOAD_LEN;
810 	static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
811 	struct iphdr *iph = (struct iphdr *)(buf + ETH_HLEN);
812 	struct ipv6hdr *ip6h = (struct ipv6hdr *)(buf + ETH_HLEN);
813 
814 	create_packet(buf, 0, 0, PAYLOAD_LEN, 0);
815 	write_packet(fd, buf, pkt_size, daddr);
816 
817 	create_packet(buf, PAYLOAD_LEN, 0, PAYLOAD_LEN, 0);
818 	if (proto == PF_INET) {
819 		iph->tos = 1;
820 		iph->check = 0;
821 		iph->check = checksum_fold(iph, sizeof(struct iphdr), 0);
822 	} else if (proto == PF_INET6) {
823 		ip6h->priority = 0xf;
824 	}
825 	write_packet(fd, buf, pkt_size, daddr);
826 }
827 
828 /* Packets with different ECN don't coalesce.*/
send_changed_ECN(int fd,struct sockaddr_ll * daddr)829 static void send_changed_ECN(int fd, struct sockaddr_ll *daddr)
830 {
831 	int pkt_size = total_hdr_len + PAYLOAD_LEN;
832 	static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
833 	struct iphdr *iph = (struct iphdr *)(buf + ETH_HLEN);
834 
835 	create_packet(buf, 0, 0, PAYLOAD_LEN, 0);
836 	write_packet(fd, buf, pkt_size, daddr);
837 
838 	create_packet(buf, PAYLOAD_LEN, 0, PAYLOAD_LEN, 0);
839 	if (proto == PF_INET) {
840 		buf[ETH_HLEN + 1] ^= 0x2; // ECN set to 10
841 		iph->check = 0;
842 		iph->check = checksum_fold(iph, sizeof(struct iphdr), 0);
843 	} else {
844 		buf[ETH_HLEN + 1] ^= 0x20; // ECN set to 10
845 	}
846 	write_packet(fd, buf, pkt_size, daddr);
847 }
848 
849 /* IPv6 fragments and packets with extensions don't coalesce.*/
send_fragment6(int fd,struct sockaddr_ll * daddr)850 static void send_fragment6(int fd, struct sockaddr_ll *daddr)
851 {
852 	static char buf[MAX_HDR_LEN + PAYLOAD_LEN];
853 	static char extpkt[MAX_HDR_LEN + PAYLOAD_LEN +
854 			   sizeof(struct ip6_frag)];
855 	struct ipv6hdr *ip6h = (struct ipv6hdr *)(buf + ETH_HLEN);
856 	struct ip6_frag *frag = (void *)(extpkt + tcp_offset);
857 	int extlen = sizeof(struct ip6_frag);
858 	int bufpkt_len = total_hdr_len + PAYLOAD_LEN;
859 	int extpkt_len = bufpkt_len + extlen;
860 	int i;
861 
862 	for (i = 0; i < 2; i++) {
863 		create_packet(buf, PAYLOAD_LEN * i, 0, PAYLOAD_LEN, 0);
864 		write_packet(fd, buf, bufpkt_len, daddr);
865 	}
866 	sleep(1);
867 	create_packet(buf, PAYLOAD_LEN * 2, 0, PAYLOAD_LEN, 0);
868 	memset(extpkt, 0, extpkt_len);
869 
870 	ip6h->nexthdr = IPPROTO_FRAGMENT;
871 	ip6h->payload_len = htons(ntohs(ip6h->payload_len) + extlen);
872 	frag->ip6f_nxt = IPPROTO_TCP;
873 
874 	memcpy(extpkt, buf, tcp_offset);
875 	memcpy(extpkt + tcp_offset + extlen, buf + tcp_offset,
876 	       sizeof(struct tcphdr) + PAYLOAD_LEN);
877 	write_packet(fd, extpkt, extpkt_len, daddr);
878 
879 	create_packet(buf, PAYLOAD_LEN * 3, 0, PAYLOAD_LEN, 0);
880 	write_packet(fd, buf, bufpkt_len, daddr);
881 }
882 
bind_packetsocket(int fd)883 static void bind_packetsocket(int fd)
884 {
885 	struct sockaddr_ll daddr = {};
886 
887 	daddr.sll_family = AF_PACKET;
888 	daddr.sll_protocol = ethhdr_proto;
889 	daddr.sll_ifindex = if_nametoindex(ifname);
890 	if (daddr.sll_ifindex == 0)
891 		error(1, errno, "if_nametoindex");
892 
893 	if (bind(fd, (void *)&daddr, sizeof(daddr)) < 0)
894 		error(1, errno, "could not bind socket");
895 }
896 
set_timeout(int fd)897 static void set_timeout(int fd)
898 {
899 	struct timeval timeout;
900 
901 	timeout.tv_sec = 3;
902 	timeout.tv_usec = 0;
903 	if (setsockopt(fd, SOL_SOCKET, SO_RCVTIMEO, (char *)&timeout,
904 		       sizeof(timeout)) < 0)
905 		error(1, errno, "cannot set timeout, setsockopt failed");
906 }
907 
check_recv_pkts(int fd,int * correct_payload,int correct_num_pkts)908 static void check_recv_pkts(int fd, int *correct_payload,
909 			    int correct_num_pkts)
910 {
911 	static char buffer[IP_MAXPACKET + ETH_HLEN + 1];
912 	struct iphdr *iph = (struct iphdr *)(buffer + ETH_HLEN);
913 	struct ipv6hdr *ip6h = (struct ipv6hdr *)(buffer + ETH_HLEN);
914 	struct tcphdr *tcph;
915 	bool bad_packet = false;
916 	int tcp_ext_len = 0;
917 	int ip_ext_len = 0;
918 	int pkt_size = -1;
919 	int data_len = 0;
920 	int num_pkt = 0;
921 	int i;
922 
923 	vlog("Expected {");
924 	for (i = 0; i < correct_num_pkts; i++)
925 		vlog("%d ", correct_payload[i]);
926 	vlog("}, Total %d packets\nReceived {", correct_num_pkts);
927 
928 	while (1) {
929 		ip_ext_len = 0;
930 		pkt_size = recv(fd, buffer, IP_MAXPACKET + ETH_HLEN + 1, 0);
931 		if (pkt_size < 0)
932 			error(1, errno, "could not receive");
933 
934 		if (iph->version == 4)
935 			ip_ext_len = (iph->ihl - 5) * 4;
936 		else if (ip6h->version == 6 && ip6h->nexthdr != IPPROTO_TCP)
937 			ip_ext_len = MIN_EXTHDR_SIZE;
938 
939 		tcph = (struct tcphdr *)(buffer + tcp_offset + ip_ext_len);
940 
941 		if (tcph->fin)
942 			break;
943 
944 		tcp_ext_len = (tcph->doff - 5) * 4;
945 		data_len = pkt_size - total_hdr_len - tcp_ext_len - ip_ext_len;
946 		/* Min ethernet frame payload is 46(ETH_ZLEN - ETH_HLEN) by RFC 802.3.
947 		 * Ipv4/tcp packets without at least 6 bytes of data will be padded.
948 		 * Packet sockets are protocol agnostic, and will not trim the padding.
949 		 */
950 		if (pkt_size == ETH_ZLEN && iph->version == 4) {
951 			data_len = ntohs(iph->tot_len)
952 				- sizeof(struct tcphdr) - sizeof(struct iphdr);
953 		}
954 		vlog("%d ", data_len);
955 		if (data_len != correct_payload[num_pkt]) {
956 			vlog("[!=%d]", correct_payload[num_pkt]);
957 			bad_packet = true;
958 		}
959 		num_pkt++;
960 	}
961 	vlog("}, Total %d packets.\n", num_pkt);
962 	if (num_pkt != correct_num_pkts)
963 		error(1, 0, "incorrect number of packets");
964 	if (bad_packet)
965 		error(1, 0, "incorrect packet geometry");
966 
967 	printf("Test succeeded\n\n");
968 }
969 
gro_sender(void)970 static void gro_sender(void)
971 {
972 	static char fin_pkt[MAX_HDR_LEN];
973 	struct sockaddr_ll daddr = {};
974 	int txfd = -1;
975 
976 	txfd = socket(PF_PACKET, SOCK_RAW, IPPROTO_RAW);
977 	if (txfd < 0)
978 		error(1, errno, "socket creation");
979 
980 	memset(&daddr, 0, sizeof(daddr));
981 	daddr.sll_ifindex = if_nametoindex(ifname);
982 	if (daddr.sll_ifindex == 0)
983 		error(1, errno, "if_nametoindex");
984 	daddr.sll_family = AF_PACKET;
985 	memcpy(daddr.sll_addr, dst_mac, ETH_ALEN);
986 	daddr.sll_halen = ETH_ALEN;
987 	create_packet(fin_pkt, PAYLOAD_LEN * 2, 0, 0, 1);
988 
989 	if (strcmp(testname, "data") == 0) {
990 		send_data_pkts(txfd, &daddr, PAYLOAD_LEN, PAYLOAD_LEN);
991 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
992 
993 		send_data_pkts(txfd, &daddr, PAYLOAD_LEN, PAYLOAD_LEN / 2);
994 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
995 
996 		send_data_pkts(txfd, &daddr, PAYLOAD_LEN / 2, PAYLOAD_LEN);
997 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
998 	} else if (strcmp(testname, "ack") == 0) {
999 		send_ack(txfd, &daddr);
1000 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1001 	} else if (strcmp(testname, "flags") == 0) {
1002 		send_flags(txfd, &daddr, 1, 0, 0, 0);
1003 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1004 
1005 		send_flags(txfd, &daddr, 0, 1, 0, 0);
1006 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1007 
1008 		send_flags(txfd, &daddr, 0, 0, 1, 0);
1009 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1010 
1011 		send_flags(txfd, &daddr, 0, 0, 0, 1);
1012 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1013 	} else if (strcmp(testname, "tcp") == 0) {
1014 		send_changed_checksum(txfd, &daddr);
1015 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1016 
1017 		send_changed_seq(txfd, &daddr);
1018 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1019 
1020 		send_changed_ts(txfd, &daddr);
1021 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1022 
1023 		send_diff_opt(txfd, &daddr);
1024 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1025 	} else if (strcmp(testname, "ip") == 0) {
1026 		send_changed_ECN(txfd, &daddr);
1027 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1028 
1029 		send_changed_tos(txfd, &daddr);
1030 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1031 		if (proto == PF_INET) {
1032 			/* Modified packets may be received out of order.
1033 			 * Sleep function added to enforce test boundaries
1034 			 * so that fin pkts are not received prior to other pkts.
1035 			 */
1036 			sleep(1);
1037 			send_changed_ttl(txfd, &daddr);
1038 			write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1039 
1040 			sleep(1);
1041 			send_ip_options(txfd, &daddr);
1042 			sleep(1);
1043 			write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1044 
1045 			sleep(1);
1046 			send_fragment4(txfd, &daddr);
1047 			sleep(1);
1048 			write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1049 
1050 			test_flush_id(txfd, &daddr, fin_pkt);
1051 		} else if (proto == PF_INET6) {
1052 			sleep(1);
1053 			send_fragment6(txfd, &daddr);
1054 			sleep(1);
1055 			write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1056 
1057 			sleep(1);
1058 			/* send IPv6 packets with ext header with same payload */
1059 			send_ipv6_exthdr(txfd, &daddr, EXT_PAYLOAD_1, EXT_PAYLOAD_1);
1060 			sleep(1);
1061 			write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1062 
1063 			sleep(1);
1064 			/* send IPv6 packets with ext header with different payload */
1065 			send_ipv6_exthdr(txfd, &daddr, EXT_PAYLOAD_1, EXT_PAYLOAD_2);
1066 			sleep(1);
1067 			write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1068 		}
1069 	} else if (strcmp(testname, "large") == 0) {
1070 		/* 20 is the difference between min iphdr size
1071 		 * and min ipv6hdr size. Like MAX_HDR_SIZE,
1072 		 * MAX_PAYLOAD is defined with the larger header of the two.
1073 		 */
1074 		int offset = proto == PF_INET ? 20 : 0;
1075 		int remainder = (MAX_PAYLOAD + offset) % MSS;
1076 
1077 		send_large(txfd, &daddr, remainder);
1078 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1079 
1080 		send_large(txfd, &daddr, remainder + 1);
1081 		write_packet(txfd, fin_pkt, total_hdr_len, &daddr);
1082 	} else {
1083 		error(1, 0, "Unknown testcase");
1084 	}
1085 
1086 	if (close(txfd))
1087 		error(1, errno, "socket close");
1088 }
1089 
gro_receiver(void)1090 static void gro_receiver(void)
1091 {
1092 	static int correct_payload[NUM_PACKETS];
1093 	int rxfd = -1;
1094 
1095 	rxfd = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_NONE));
1096 	if (rxfd < 0)
1097 		error(1, 0, "socket creation");
1098 	setup_sock_filter(rxfd);
1099 	set_timeout(rxfd);
1100 	bind_packetsocket(rxfd);
1101 
1102 	memset(correct_payload, 0, sizeof(correct_payload));
1103 
1104 	if (strcmp(testname, "data") == 0) {
1105 		printf("pure data packet of same size: ");
1106 		correct_payload[0] = PAYLOAD_LEN * 2;
1107 		check_recv_pkts(rxfd, correct_payload, 1);
1108 
1109 		printf("large data packets followed by a smaller one: ");
1110 		correct_payload[0] = PAYLOAD_LEN * 1.5;
1111 		check_recv_pkts(rxfd, correct_payload, 1);
1112 
1113 		printf("small data packets followed by a larger one: ");
1114 		correct_payload[0] = PAYLOAD_LEN / 2;
1115 		correct_payload[1] = PAYLOAD_LEN;
1116 		check_recv_pkts(rxfd, correct_payload, 2);
1117 	} else if (strcmp(testname, "ack") == 0) {
1118 		printf("duplicate ack and pure ack: ");
1119 		check_recv_pkts(rxfd, correct_payload, 3);
1120 	} else if (strcmp(testname, "flags") == 0) {
1121 		correct_payload[0] = PAYLOAD_LEN * 3;
1122 		correct_payload[1] = PAYLOAD_LEN * 2;
1123 
1124 		printf("psh flag ends coalescing: ");
1125 		check_recv_pkts(rxfd, correct_payload, 2);
1126 
1127 		correct_payload[0] = PAYLOAD_LEN * 2;
1128 		correct_payload[1] = 0;
1129 		correct_payload[2] = PAYLOAD_LEN * 2;
1130 		printf("syn flag ends coalescing: ");
1131 		check_recv_pkts(rxfd, correct_payload, 3);
1132 
1133 		printf("rst flag ends coalescing: ");
1134 		check_recv_pkts(rxfd, correct_payload, 3);
1135 
1136 		printf("urg flag ends coalescing: ");
1137 		check_recv_pkts(rxfd, correct_payload, 3);
1138 	} else if (strcmp(testname, "tcp") == 0) {
1139 		correct_payload[0] = PAYLOAD_LEN;
1140 		correct_payload[1] = PAYLOAD_LEN;
1141 		correct_payload[2] = PAYLOAD_LEN;
1142 		correct_payload[3] = PAYLOAD_LEN;
1143 
1144 		printf("changed checksum does not coalesce: ");
1145 		check_recv_pkts(rxfd, correct_payload, 2);
1146 
1147 		printf("Wrong Seq number doesn't coalesce: ");
1148 		check_recv_pkts(rxfd, correct_payload, 2);
1149 
1150 		printf("Different timestamp doesn't coalesce: ");
1151 		correct_payload[0] = PAYLOAD_LEN * 2;
1152 		check_recv_pkts(rxfd, correct_payload, 4);
1153 
1154 		printf("Different options doesn't coalesce: ");
1155 		correct_payload[0] = PAYLOAD_LEN * 2;
1156 		check_recv_pkts(rxfd, correct_payload, 2);
1157 	} else if (strcmp(testname, "ip") == 0) {
1158 		correct_payload[0] = PAYLOAD_LEN;
1159 		correct_payload[1] = PAYLOAD_LEN;
1160 
1161 		printf("different ECN doesn't coalesce: ");
1162 		check_recv_pkts(rxfd, correct_payload, 2);
1163 
1164 		printf("different tos doesn't coalesce: ");
1165 		check_recv_pkts(rxfd, correct_payload, 2);
1166 
1167 		if (proto == PF_INET) {
1168 			printf("different ttl doesn't coalesce: ");
1169 			check_recv_pkts(rxfd, correct_payload, 2);
1170 
1171 			printf("ip options doesn't coalesce: ");
1172 			correct_payload[2] = PAYLOAD_LEN;
1173 			check_recv_pkts(rxfd, correct_payload, 3);
1174 
1175 			printf("fragmented ip4 doesn't coalesce: ");
1176 			check_recv_pkts(rxfd, correct_payload, 2);
1177 
1178 			/* is_atomic checks */
1179 			printf("DF=1, Incrementing - should coalesce: ");
1180 			correct_payload[0] = PAYLOAD_LEN * 2;
1181 			check_recv_pkts(rxfd, correct_payload, 1);
1182 
1183 			printf("DF=1, Fixed - should coalesce: ");
1184 			correct_payload[0] = PAYLOAD_LEN * 2;
1185 			check_recv_pkts(rxfd, correct_payload, 1);
1186 
1187 			printf("DF=0, Incrementing - should coalesce: ");
1188 			correct_payload[0] = PAYLOAD_LEN * 2;
1189 			check_recv_pkts(rxfd, correct_payload, 1);
1190 
1191 			printf("DF=0, Fixed - should not coalesce: ");
1192 			correct_payload[0] = PAYLOAD_LEN;
1193 			correct_payload[1] = PAYLOAD_LEN;
1194 			check_recv_pkts(rxfd, correct_payload, 2);
1195 
1196 			printf("DF=1, 2 Incrementing and one fixed - should coalesce only first 2 packets: ");
1197 			correct_payload[0] = PAYLOAD_LEN * 2;
1198 			correct_payload[1] = PAYLOAD_LEN;
1199 			check_recv_pkts(rxfd, correct_payload, 2);
1200 
1201 			printf("DF=1, 2 Fixed and one incrementing - should coalesce only first 2 packets: ");
1202 			correct_payload[0] = PAYLOAD_LEN * 2;
1203 			correct_payload[1] = PAYLOAD_LEN;
1204 			check_recv_pkts(rxfd, correct_payload, 2);
1205 		} else if (proto == PF_INET6) {
1206 			/* GRO doesn't check for ipv6 hop limit when flushing.
1207 			 * Hence no corresponding test to the ipv4 case.
1208 			 */
1209 			printf("fragmented ip6 doesn't coalesce: ");
1210 			correct_payload[0] = PAYLOAD_LEN * 2;
1211 			correct_payload[1] = PAYLOAD_LEN;
1212 			correct_payload[2] = PAYLOAD_LEN;
1213 			check_recv_pkts(rxfd, correct_payload, 3);
1214 
1215 			printf("ipv6 with ext header does coalesce: ");
1216 			correct_payload[0] = PAYLOAD_LEN * 2;
1217 			check_recv_pkts(rxfd, correct_payload, 1);
1218 
1219 			printf("ipv6 with ext header with different payloads doesn't coalesce: ");
1220 			correct_payload[0] = PAYLOAD_LEN;
1221 			correct_payload[1] = PAYLOAD_LEN;
1222 			check_recv_pkts(rxfd, correct_payload, 2);
1223 		}
1224 	} else if (strcmp(testname, "large") == 0) {
1225 		int offset = proto == PF_INET ? 20 : 0;
1226 		int remainder = (MAX_PAYLOAD + offset) % MSS;
1227 
1228 		correct_payload[0] = (MAX_PAYLOAD + offset);
1229 		correct_payload[1] = remainder;
1230 		printf("Shouldn't coalesce if exceed IP max pkt size: ");
1231 		check_recv_pkts(rxfd, correct_payload, 2);
1232 
1233 		/* last segment sent individually, doesn't start new segment */
1234 		correct_payload[0] = correct_payload[0] - remainder;
1235 		correct_payload[1] = remainder + 1;
1236 		correct_payload[2] = remainder + 1;
1237 		check_recv_pkts(rxfd, correct_payload, 3);
1238 	} else {
1239 		error(1, 0, "Test case error, should never trigger");
1240 	}
1241 
1242 	if (close(rxfd))
1243 		error(1, 0, "socket close");
1244 }
1245 
parse_args(int argc,char ** argv)1246 static void parse_args(int argc, char **argv)
1247 {
1248 	static const struct option opts[] = {
1249 		{ "daddr", required_argument, NULL, 'd' },
1250 		{ "dmac", required_argument, NULL, 'D' },
1251 		{ "iface", required_argument, NULL, 'i' },
1252 		{ "ipv4", no_argument, NULL, '4' },
1253 		{ "ipv6", no_argument, NULL, '6' },
1254 		{ "rx", no_argument, NULL, 'r' },
1255 		{ "saddr", required_argument, NULL, 's' },
1256 		{ "smac", required_argument, NULL, 'S' },
1257 		{ "test", required_argument, NULL, 't' },
1258 		{ "verbose", no_argument, NULL, 'v' },
1259 		{ 0, 0, 0, 0 }
1260 	};
1261 	int c;
1262 
1263 	while ((c = getopt_long(argc, argv, "46d:D:i:rs:S:t:v", opts, NULL)) != -1) {
1264 		switch (c) {
1265 		case '4':
1266 			proto = PF_INET;
1267 			ethhdr_proto = htons(ETH_P_IP);
1268 			break;
1269 		case '6':
1270 			proto = PF_INET6;
1271 			ethhdr_proto = htons(ETH_P_IPV6);
1272 			break;
1273 		case 'd':
1274 			addr4_dst = addr6_dst = optarg;
1275 			break;
1276 		case 'D':
1277 			dmac = optarg;
1278 			break;
1279 		case 'i':
1280 			ifname = optarg;
1281 			break;
1282 		case 'r':
1283 			tx_socket = false;
1284 			break;
1285 		case 's':
1286 			addr4_src = addr6_src = optarg;
1287 			break;
1288 		case 'S':
1289 			smac = optarg;
1290 			break;
1291 		case 't':
1292 			testname = optarg;
1293 			break;
1294 		case 'v':
1295 			verbose = true;
1296 			break;
1297 		default:
1298 			error(1, 0, "%s invalid option %c\n", __func__, c);
1299 			break;
1300 		}
1301 	}
1302 }
1303 
main(int argc,char ** argv)1304 int main(int argc, char **argv)
1305 {
1306 	parse_args(argc, argv);
1307 
1308 	if (proto == PF_INET) {
1309 		tcp_offset = ETH_HLEN + sizeof(struct iphdr);
1310 		total_hdr_len = tcp_offset + sizeof(struct tcphdr);
1311 	} else if (proto == PF_INET6) {
1312 		tcp_offset = ETH_HLEN + sizeof(struct ipv6hdr);
1313 		total_hdr_len = MAX_HDR_LEN;
1314 	} else {
1315 		error(1, 0, "Protocol family is not ipv4 or ipv6");
1316 	}
1317 
1318 	read_MAC(src_mac, smac);
1319 	read_MAC(dst_mac, dmac);
1320 
1321 	if (tx_socket)
1322 		gro_sender();
1323 	else
1324 		gro_receiver();
1325 
1326 	fprintf(stderr, "Gro::%s test passed.\n", testname);
1327 	return 0;
1328 }
1329